Difference between revisions of "PTP/meetings/February 2013"

Latest revision as of 16:29, 28 May 2013

February 2013 Meeting

Date: February 12, 2013 Time: 1:00 PM EDT

Call in information:

Toll Free (US): 888-426-6840
Toll Free (Germany): 0800-000-1018
Toll Free (UK): 0800-368-0638
Passcode: 2221402

Agenda

1. Overall project, releases, etc.
   • Next builds, releases
     • 6.0.x Juno - today is last build day for SR2
       • see Juno Planning page
     • 7.0.0 Kepler
       • see Kepler Planning page
       • M5 was last week
       • M6+3 is 3/27
   • Download stats (see below)
     • 263,891 total Juno Parallel package to 2/11
2. Status Reports
   • Bug status/triage - still need to schedule
     • CDT remote include fix
   • PTP Core/debug
   • RDT, Synchronized Project, Remote status
   • Target System Configurations (previously known as Resource Managers)
   • Monitoring
   • PLDT
   • ETFw, other tool integrations
   • Photran
   • Docs, Help, N&N
   • Other PTP contributions
3. Education and Outreach
   • Monthly PTP User call - last one was Wed Jan 23 - see PTP User Meetings - website needs to be updated
     • Next one is Feb 27
   • Tutorials, BOFs, other outreach opportunities
4. Other Business
   • EclipseCon is March 25-28 in Boston
5. Next meeting - Tuesday March 12

Attendees

Greg Watson, John Eblen, Wyatt Spear, Dave Wootton, Carsten Karbach, Chris Navarro, Brian Watt, Jeff Overbey, Jay Alameda

Minutes

1. Overall project, releases, etc.
   • Next builds, releases
     • 6.0.x Juno - today is last build day for SR2
       • see Juno Planning page 
       • Main issue – is Juno regression problem, in terms of next build – SR2 – is last build day for this is today, RC4 build.  And – have a problem that people are seeing, this issue with RSE initialization – for Juno SR2, this is the main issue.  Greg has SR2 – RC2 and RC3 – installed on Mac, and if he goes and create a new synchronized project, RSE init is hanging in RC3,but not in RC2.  Definitely a regression here.
         • David D. has come back and said there was a change they made to fix initialization for CDT, he suggested using RC4a – when he looked at RC4A, looks the same as what using in RC3 –
           Regression the other way, were not seeing this issue in RC2, can’t see how RC4 will help – presumably the same plugin.
         • They have fixed an issue for someone else, have a bit of conflict here – between someone else and us – need to figure out how to resolve this.
           Option for us – disable the RSE, disable the RSE plugins out of the parallel package –
         • Last commit in PTP – was in RSE –spawner/miner:
         • The miner – has to do with remote projects, and this is the protocol to communicate between client and server –
           this had to do with dstore, and rse package –
         • Only solution is to disable RSE –
           Synchronized projects will use remote tools –this would be a punt, as
           This would be a way to remove the bridge – between synchronized projects, and the RSE initialization
           
         • N.B. update late Tuesday: David D has a fix; Greg has verified it seems to fix PTP hang; See https://bugs.eclipse.org/bugs/show_bug.cgi?id=400345
     • 7.0.0 Kepler
       • see Kepler Planning page
       • M5 was last week
       • M6+3 is 3/27
       • In terms of Kepler, on M5 now – still have a number of builds – M6 and M7 to go yet –
         And RC builds start in May
   • Download stats (see below)
     • 263,891 total Juno Parallel package to 2/11
2. Status Reports
   • Bug status/triage - still need to schedule
     • CDT remote include fix
     • CDT remote include fix – has been committed to master branch of CDT, which means Kepler, it won’t make it to Juno. There are some issues with this fix – if you have your project configured to use remote includes, and indexer on – every time indexer runs, will try to make connections to remote machine – if have networking issues, indexer could get stuck – tried to include mechanisms to cancel indexer in those situations – but CDT didn’t like them, and they didn’t work anyway – as jsch ssh code is not cancelable – can add all cancel calls you like, but if you drill down into jsch code, there is a read system call, with no time outs or anything, there are no timeouts in PTP – so this is a blocking call.  Don’t know of an easy way to resolve this.  It can cause some issues.
       
   • PTP Core/debug
   • RDT, Synchronized Project, Remote status
     • John Eblen – anything on the synchronized projects, noticed reverted the changes – need to revisit for Kepler – part of this was specifying git location
       And if there is some way to automate the verification of the executable – could take a long time, could add checkbox to make verification optional or not –
       Could discuss this some more on the bug –
       Env mgt: want to make per-configuration –
       And had a question about new remote scanner/discovery stuff – noticed you added this, seems to work fine, but it is not enabled by default, is it possible to make it default –
       Disabled std scanner/discovery, and enabled new one – seemed to discover system include paths on the remote machine, still  more to do with integrating with EMS as well –
       May want to be able to change say compiler, with EMS, and rerun scanner discovery – in this scenario – tool chain is a bit irrelevant, EMS is actually choosing toolchain on remote system ,not the one you chose in the wizard.  Thinking a little bit if you don’t want any tool chain, or we have a special toolchain, called “remote toolchain” or something that gets automatically created by EMS system.  Have you thought about best way to go here?
       Not yet –thought about getting rid of the big list of toolchains,
       Some cases where want to use one – where not using env mgt, but if using env mgt, don’t want to pick a toolchain –
       Will open a bug on this –
       
   • Target System Configurations (previously known as Resource Managers): Api changes to jaxb – result of merging of jaxb and eftw – will be committing this – some is fairly significant.  Most significant thing is change the way you submit interactive jobs – batch interactive jobs – this is something that has been annoying Greg for some time, if are submitting a pbs –I job, click submit – this will run the job, when the job starts running –
   • Then, when click run button on the IDE, would submit the command.  This had some state – that would block other submissions to job scheduler –Change is – that now can submit as many jobs as you want, global state is gone – do this through the monitoring view – right click job, and another item that says “resubmit” – so it will be job-specific, not a global thing. May have broken some things in the process, need some careful testing to make sure job submission is still working.
   • Monitoring
     • Carsten – on monitoring –
       Working on LML, discussing on how to split LML into layout and data part, to have LML layout be in target system configuration – this has now been done, so have LML requests and LML responses – documented this in bug 360435.  Wanted to know how you think about it, and now – with splitting the LML layout, still need about 50 new classes generated from xml schema, would be great if could put special layout config into target system configuration –
       Want to extend the target configuration schema to add layout elements? Correct.  Already have monitoring element in target system configuration – Greg agrees, would like to create a patch with the new schema and also – generate the classes as well?  Carsten has generated the patch, can put the new classes into the same package as the parent classes.  Only commit to master for Kepler.  In bug, wanted to know where to commit the change, as big change to LML schema – and also need to change LML core plugin – as LML schema changes, and jaxb classes of LML change too –
       If update patch, then Greg will look at it past SR2 –
       Brian – doing some reorganization, how much impact to LML code? Not much, mostly jaxb – some tiny places where package name has changed –
       Testing on juqeen and judge, did some minor improvements on LML –
       Also working to get it working properly on darpa system, hope to get Beth to test on the darpa machine – This seems to be a LoadLeveler issue – discrepancy between node names – if fully qualified or not, certain commands return fully qualified hostname, others, only first part of hostname –
       This may have been in a version change in LoadLeveler which changed how this is working –
       Is your patch covering the old case?  Or all 3 cases: fully qualified in all cases, mix of fully qualified and not fully qualified, or #3 – not fully qualified –
       Does Dave still have a LoadLeveler machine set up? Had one – last time worked with it, had some trouble with it –
       Need to make sure it works with old versions of LoadLeveler –
       
       
   • PLDT
   • ETFw, other tool integrations: Chris and Wyatt – and Brian – Brian: idea is to provide extension point to jaxb code, to allow custom widget to be defined – widget class, and update model – 2 classes – can put your own code in here to set this up.  Modified env config to use this capability, new package out there to allow this – hope we don’t have to continue to add more low-level capability to rm jaxb code – but just extend
     • Chris – based on this, made some changes –  also now have initialization and cleanup commands – Did some code cleanup too – made it a lot cleaner – Rm validation objects that are created – resource manager specific –o Once greg commits Brian’s patches from the bugzilla, then Chris needs to apply patches, and Greg review, and we can then move forward.
     • Greg – we are getting pretty close, work Brian and Chris has been doing, have most of the pieces to bring this all together.  Wyatt – do you have any comments on this – it is all really exciting, can I get name of the plugin with the widget extension point?  Plugin is org.eclipse.ptp.emf.control.ui is the new custom widget, with the 2 classes – it is in the patch.  And, optional tar file added to the bugzilla, to add a most basic widget.
       ETFW – in transition to jaxb, not doing a whole lot, mostly debugging – from sound of this, sounds like we’ll be ready to support etfw – Chris is thinking we can phase the transition – keep both in right now, mostly bug fixing and testing for the relase –
       
     • N.B. - David D has a fix, Greg says it works to fix PTP hang. See https://bugs.eclipse.org/bugs/show_bug.cgi?id=400345
   • Photran
     • Jeff - Bit of activity on fortran stuff – anything happening there, and new job? Need to find funding here or something,
       Have breakpoint bug fixed – thanks to the intel guys for helping -  and Louis from TechX putting patches in for Kepler, all small – CQ deadline was Friday, but not hard and fast, it is a positioning in the queue –
       
   • Docs, Help, N&N
   • Other PTP contributions
   • Dave Wooton – on LSF – have first pass – with jaxb, list of options, fill in the values –
     Started improving on this for Kepler, would like to have this up and running by Mid-March – try out SR2 version, give it a try –
     SR2 version will figure out queue names – dynamically, and works with LML code – in Kepler, will have nice simplified user interface with tabs and stuff –
     Testing of monitoring – limited by Dave, did some testing on this – and mini-cluster, some simple serial jobs, and can also cancel jobs from monitoring view –
     Carsten can test on system he has access on –
     Can bring out PTP bug fix release –
     Can test on Yellowstone as well –
     
3. Education and Outreach
   • Monthly PTP User call - last one was Wed Jan 23 - see PTP User Meetings - website needs to be updated
     • Next one is Feb 27
   • Tutorials, BOFs, other outreach opportunities
4. Other Business
   • EclipseCon is March 25-28 in Boston
     
   • After Kepler: Luna…
5. Next meeting - Tuesday March 12

Parallel Package Downloads

Download numbers, by month, of 'Eclipse for Parallel Application Developers' available on http://eclipse.org/downloads. Each release (e.g. R,SR1,SR2) the counts are set to 0 and restarted.

February* downloads as of February 11.

<< Previous month stats | Next month stats >>